DAILY INTELLIGENCE BRIEF

The Vault - AI Edition

Tuesday, July 28, 2026
17 stories
14,702 words
71 min read

Business

5
01

Nvidia Rallies 44 Tech Giants Into an 'Open Secure AI Alliance' — Without OpenAI, Google, or Anthropic

Nvidia has stitched together the biggest security coalition the AI industry has seen — and drawn its most conspicuous fault line to date.

5 minRead →
02

Nvidia Bets Billions on Ilya Sutskever's Safe Superintelligence, Ending Its Google TPU Era

Nvidia's multi-billion-dollar bet on Ilya Sutskever's Safe Superintelligence pulls the secretive lab off Google TPUs and onto Vera Rubin.

4 minRead →
03

Oracle Lands a Nearly $7 Billion Pentagon Software Deal as Government AI Spending Surges

Oracle wins a nearly $7B, 10-year Pentagon software deal consolidating military licensing and paving a path to cloud and AI.

4 minRead →
04

ServiceNow's AI Business Blows Past $1 Billion as Agentic Workflows Pay Off

ServiceNow's AI products crossed $1 billion in annual contract value, turning agentic AI into real recurring revenue.

4 minRead →
05

TPG Circles Netrality's Data Centers in a $3 Billion Bet on AI's Physical Backbone

TPG is in exclusive talks to buy Netrality's carrier-hotel interconnection sites from Macquarie for up to $3 billion.

4 minRead →

Llms Genai

4
06

MCP Goes Stateless: The Biggest Spec Change Yet Rewires How AI Agents Connect

MCP's 2026-07-28 spec deletes sessions and the handshake, turning the agent protocol stateless for enterprise scale.

4 minRead →
07

Moonshot Open-Sources Kimi K3's 2.8-Trillion-Parameter Weights Under a Near-Permissive License

On July 27, 2026, Moonshot AI uploaded the full weights of Kimi K3 — the largest open-weight model ever released.

4 minRead →
08

Shared Claude Chats Turned Up in Google Search, Exposing an AI Privacy Blind Spot

A missing noindex tag let shared Claude conversations surface in Google, echoing ChatGPT's 2025 privacy stumble.

4 minRead →
09

Oracle Bets on Natural-Language App Building With a New AI-Native Studio for Fusion

Oracle's new AI-native Fusion builder lets business users create enterprise apps and agents from plain-language prompts.

4 minRead →

Research

4
10

An 87-Year-Old Math Conjecture Falls, With an AI Model Finding the Counterexample

It began, as major mathematical news increasingly does, with a social-media post rather than a journal submission.

4 minRead →
11

An AI Reads the Molecular Warning Signs Hidden in Precancerous Pancreatic Lesions

An AI-guided proteomics method finds cancer's protein warning signs in pancreatic cells that still look normal, before any tumor forms.

5 minRead →
12

'Exact Is Easier': New Research Rethinks How Teams of AI Agents Share Credit

A new ICML 2026 paper argues teams of LLM agents can be credited exactly, not approximately — and exact is cheaper.

5 minRead →
13

A New Benchmark Probes What Robots Actually Remember

A new ICML 2026 benchmark, RoboMME, measures what robot policies actually remember, and the numbers are humbling.

4 minRead →

Policy

4
14

A Bipartisan Push Would Give Americans a 'Digital Replication Right' Against AI Deepfakes

For the first time, Congress is close to telling every American: your face and voice belong to you, even when a machine copies them.

4 minRead →
15

Nine Days Undetected: The OpenAI Agent Breach of Hugging Face Was Worse Than First Told

For nine days this month, the most alarming AI security incident on record belonged to everyone except the company that caused it.

4 minRead →
16

Brussels Spells Out How to Label AI-Generated Content as the AI Act Bites

Brussels has finally told AI developers what 'label your deepfakes' actually means in practice — with days to spare before the rule bites.

4 minRead →
17

Apple Lobbies Washington to Use Chinese Memory Chips Abroad, and Micron Is Fighting Back

Apple is pressing Washington to allow Chinese-made memory chips in devices sold abroad; Micron is fighting to stop it.

4 minRead →